Tonight, Google engineer and artist Mike Tyka discusses with Gray Area founder Josette Melchor the creative possibilities afforded neural networks and the impact the DeepDream project has had.
When DeepDream was released in 2015, the Google program became an overnight sensation, garnering coverage from magazines such as Wired, the Atlantic, and Scientific American. Ordinary photos processed through the algorithm transformed into LSD-reminiscent hallucinations: clouds writhed with dogs, palms filled with eyes.
The program is an artificial neural network aimed at generating new images. Artificial neural networks are a biologically inspired form of computing which, unlike classical computer algorithms, aren’t programmed directly by human operators but instead learn from large amounts of example data. The software is trained with natural images from the environment to distinguish objects and parse them into high level features. At Google, their usual application is image classification and object recognition.
The DeepDream Google engineers wanted to test the extent to which a neural network had learned to recognize such features by asking the computer to describe what it saw. So they fed their program images and asked it to enhance the components it recognized. If a particular neuron tended to tag images as frogs, frogs would proliferate in the modified picture. In this way, the network was manipulated to produce new imagery, essentially “imagining” the images based on learned rules and associations.
As engineers Alexander Mordvintsev, Christopher Olah, and Mike Tyka wrote in a developer’s blog post: “The results vary quite a bit with the kind of image, because the features that are entered bias the network towards certain interpretations. For example, horizon lines tend to get filled with towers and pagodas. Rocks and trees turn into buildings. Birds and insects appear in images of leaves.”
The Google team recognized the artistic potential of Deep Dream and released the algorithm for anyone to download. In the next few days, the program went viral. Six months later, Gray Area hosted the first artificial neural network art exhibition and auctioned artworks created for thousands of dollars.